In [2]:
import tensorflow as tf
from tensorflow.keras import models, layers
import matplotlib.pyplot as plt
import numpy as np
In [3]:
####### Variables #####"

IMAGE_SIZE = 256
BATCH_SIZE = 32
RGB_CHANNELS = 3
EPOCHS = 50 #50
In [4]:
imageDataset = tf.keras.preprocessing.image_dataset_from_directory(
    "dataset/",
    shuffle=True,
    image_size= (IMAGE_SIZE, IMAGE_SIZE),
    batch_size=BATCH_SIZE
)

Found 2152 files belonging to 3 classes.
In [6]:
class_names = imageDataset.class_names
class_names
Out[6]:
['Potato___Early_blight', 'Potato___Late_blight', 'Potato___healthy']
In [7]:
len(imageDataset)
Out[7]:
68
In [8]:
for image_batch, label_batch in imageDataset.take(1):
    #print(image_batch.shape)
    #print(label_batch.numpy())
    print(image_batch[0].shape)

(256, 256, 3)
In [9]:
plt.figure(figsize=(15, 15))

for image_batch, label_batch in imageDataset.take(1):
    
    """plt.imshow(image_batch[0].numpy().astype("uint8"))
    plt.title(class_names[label_batch[0]])"""
    for i in range(12):
        ax = plt.subplot(3, 4, i+1)
        plt.imshow(image_batch[i].numpy().astype("uint8"))
        plt.title(class_names[label_batch[i]])
        plt.axis("off")
No description has been provided for this image
In [ ]:
########  Split our dataset

# 80% ==> training
# 20% ==> 10% validation, 10% test
In [10]:
train_size = 0.8
len(imageDataset)*train_size # 80% of our dataset
Out[10]:
54.400000000000006
In [11]:
train_set = imageDataset.take(54)
len(train_set)
Out[11]:
54
In [12]:
test_set = imageDataset.skip(54)
len(test_set)
Out[12]:
14
In [13]:
validation_size = 0.1
len(imageDataset)*validation_size
Out[13]:
6.800000000000001
In [14]:
validation_set = test_set.take(6)
len(validation_set)
Out[14]:
6
In [15]:
test_set = test_set.skip(len(validation_set))
len(test_set)
Out[15]:
8
In [16]:
def get_dataset_partitions_tf(ds, train_split=0.8, val_split=0.1, test_split=0.1, shuffle=True, shuffle_size=10000 ):

    ds_size = len(ds)
    if shuffle:
        ds = ds.shuffle(shuffle_size, seed=10)
        
    train_size = int(train_split * ds_size)
    val_size = int(val_split * ds_size)
    train_set = ds.take(train_size)
    val_set = ds.skip(train_size).take(val_size)
    test_set = ds.skip(train_size).skip(val_size)
    
    return train_set, validation_set, test_set
In [17]:
train_set, validation_set, test_set = get_dataset_partitions_tf(imageDataset)
In [18]:
len(train_set)
Out[18]:
54
In [ ]:
# catching and fetching
In [19]:
# read image from disk and put it in cache for next usages
train_set = train_set.cache().shuffle(1000).prefetch(buffer_size = tf.data.AUTOTUNE)
validation_set = validation_set.cache().shuffle(1000).prefetch(buffer_size = tf.data.AUTOTUNE)
test_set = test_set.cache().shuffle(1000).prefetch(buffer_size = tf.data.AUTOTUNE)
In [ ]:
# some preprocessing
In [20]:
resize_and_rescale = tf.keras.Sequential([
    layers.experimental.preprocessing.Resizing(IMAGE_SIZE, IMAGE_SIZE),
    layers.experimental.preprocessing.Rescaling(1.0/255)
])
In [ ]:
# Data augmentation
In [21]:
data_augmentation = tf.keras.Sequential([
    layers.experimental.preprocessing.RandomFlip("horizontal_and_vertical"),
    layers.experimental.preprocessing.RandomRotation(0.2),
])
In [ ]:
# Build, train and save the Model : CNN Model
In [22]:
input_shape = (BATCH_SIZE, IMAGE_SIZE, IMAGE_SIZE, RGB_CHANNELS)
nbr_classes = 3
In [23]:
model = models.Sequential([
    resize_and_rescale,
    data_augmentation,
    layers.Conv2D(32, kernel_size = (3,3), activation='relu', input_shape = input_shape ),
    layers.MaxPooling2D((2,2)),

    layers.Conv2D(64, kernel_size = (3,3), activation='relu'),
    layers.MaxPooling2D((2,2)),

    layers.Conv2D(64, kernel_size = (3,3), activation='relu'),
    layers.MaxPooling2D((2,2)),

    layers.Conv2D(64, kernel_size = (3,3), activation='relu'),
    layers.MaxPooling2D((2,2)),

    layers.Conv2D(64, kernel_size = (3,3), activation='relu'),
    layers.MaxPooling2D((2,2)),

    layers.Conv2D(64, kernel_size = (3,3), activation='relu'),
    layers.MaxPooling2D((2,2)),

    layers.Flatten(),
    layers.Dense(64, activation='relu'),
    layers.Dense(nbr_classes, activation='softmax'),
])

model.build(input_shape = input_shape)
In [24]:
model.summary()

Model: "sequential_2"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 sequential (Sequential)     (32, 256, 256, 3)         0         
                                                                 
 sequential_1 (Sequential)   (32, 256, 256, 3)         0         
                                                                 
 conv2d (Conv2D)             (32, 254, 254, 32)        896       
                                                                 
 max_pooling2d (MaxPooling2  (32, 127, 127, 32)        0         
 D)                                                              
                                                                 
 conv2d_1 (Conv2D)           (32, 125, 125, 64)        18496     
                                                                 
 max_pooling2d_1 (MaxPoolin  (32, 62, 62, 64)          0         
 g2D)                                                            
                                                                 
 conv2d_2 (Conv2D)           (32, 60, 60, 64)          36928     
                                                                 
 max_pooling2d_2 (MaxPoolin  (32, 30, 30, 64)          0         
 g2D)                                                            
                                                                 
 conv2d_3 (Conv2D)           (32, 28, 28, 64)          36928     
                                                                 
 max_pooling2d_3 (MaxPoolin  (32, 14, 14, 64)          0         
 g2D)                                                            
                                                                 
 conv2d_4 (Conv2D)           (32, 12, 12, 64)          36928     
                                                                 
 max_pooling2d_4 (MaxPoolin  (32, 6, 6, 64)            0         
 g2D)                                                            
                                                                 
 conv2d_5 (Conv2D)           (32, 4, 4, 64)            36928     
                                                                 
 max_pooling2d_5 (MaxPoolin  (32, 2, 2, 64)            0         
 g2D)                                                            
                                                                 
 flatten (Flatten)           (32, 256)                 0         
                                                                 
 dense (Dense)               (32, 64)                  16448     
                                                                 
 dense_1 (Dense)             (32, 3)                   195       
                                                                 
=================================================================
Total params: 183747 (717.76 KB)
Trainable params: 183747 (717.76 KB)
Non-trainable params: 0 (0.00 Byte)
_________________________________________________________________
In [25]:
#Compile for uptimization

model.compile(
    optimizer='adam',
    loss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
    metrics =  ['accuracy']
)
In [26]:
history = model.fit(
    train_set,
    epochs = EPOCHS,
    batch_size = BATCH_SIZE,
    verbose = 1,
    validation_data = validation_set,
)

Epoch 1/50
54/54 [==============================] - 28s 503ms/step - loss: 0.9097 - accuracy: 0.4671 - val_loss: 0.8448 - val_accuracy: 0.5469
Epoch 2/50
54/54 [==============================] - 42s 785ms/step - loss: 0.7673 - accuracy: 0.6461 - val_loss: 0.4900 - val_accuracy: 0.7917
Epoch 3/50
54/54 [==============================] - 46s 847ms/step - loss: 0.5755 - accuracy: 0.7694 - val_loss: 0.3885 - val_accuracy: 0.8125
Epoch 4/50
54/54 [==============================] - 46s 849ms/step - loss: 0.4174 - accuracy: 0.8257 - val_loss: 0.2254 - val_accuracy: 0.9271
Epoch 5/50
54/54 [==============================] - 46s 848ms/step - loss: 0.2640 - accuracy: 0.8950 - val_loss: 0.1732 - val_accuracy: 0.9427
Epoch 6/50
54/54 [==============================] - 46s 852ms/step - loss: 0.2458 - accuracy: 0.9085 - val_loss: 0.3344 - val_accuracy: 0.8542
Epoch 7/50
54/54 [==============================] - 46s 847ms/step - loss: 0.2306 - accuracy: 0.9102 - val_loss: 0.1159 - val_accuracy: 0.9688
Epoch 8/50
54/54 [==============================] - 46s 845ms/step - loss: 0.1964 - accuracy: 0.9255 - val_loss: 0.1805 - val_accuracy: 0.8958
Epoch 9/50
54/54 [==============================] - 46s 848ms/step - loss: 0.1783 - accuracy: 0.9313 - val_loss: 0.1928 - val_accuracy: 0.9271
Epoch 10/50
54/54 [==============================] - 46s 846ms/step - loss: 0.2333 - accuracy: 0.9155 - val_loss: 0.1995 - val_accuracy: 0.9062
Epoch 11/50
54/54 [==============================] - 46s 845ms/step - loss: 0.1819 - accuracy: 0.9302 - val_loss: 0.1004 - val_accuracy: 0.9583
Epoch 12/50
54/54 [==============================] - 46s 846ms/step - loss: 0.1909 - accuracy: 0.9225 - val_loss: 0.0844 - val_accuracy: 0.9844
Epoch 13/50
54/54 [==============================] - 46s 845ms/step - loss: 0.1616 - accuracy: 0.9384 - val_loss: 0.0887 - val_accuracy: 0.9583
Epoch 14/50
54/54 [==============================] - 46s 845ms/step - loss: 0.1502 - accuracy: 0.9396 - val_loss: 0.2659 - val_accuracy: 0.8854
Epoch 15/50
54/54 [==============================] - 46s 846ms/step - loss: 0.1228 - accuracy: 0.9519 - val_loss: 0.1184 - val_accuracy: 0.9479
Epoch 16/50
54/54 [==============================] - 46s 844ms/step - loss: 0.1232 - accuracy: 0.9583 - val_loss: 0.1153 - val_accuracy: 0.9635
Epoch 17/50
54/54 [==============================] - 46s 845ms/step - loss: 0.0857 - accuracy: 0.9677 - val_loss: 0.0595 - val_accuracy: 0.9688
Epoch 18/50
54/54 [==============================] - 46s 845ms/step - loss: 0.0907 - accuracy: 0.9677 - val_loss: 0.1301 - val_accuracy: 0.9323
Epoch 19/50
54/54 [==============================] - 46s 853ms/step - loss: 0.0747 - accuracy: 0.9736 - val_loss: 0.3579 - val_accuracy: 0.8854
Epoch 20/50
54/54 [==============================] - 46s 848ms/step - loss: 0.0817 - accuracy: 0.9665 - val_loss: 0.1707 - val_accuracy: 0.9219
Epoch 21/50
54/54 [==============================] - 46s 846ms/step - loss: 0.0708 - accuracy: 0.9707 - val_loss: 0.1766 - val_accuracy: 0.9427
Epoch 22/50
54/54 [==============================] - 46s 850ms/step - loss: 0.0732 - accuracy: 0.9742 - val_loss: 0.0701 - val_accuracy: 0.9688
Epoch 23/50
54/54 [==============================] - 46s 845ms/step - loss: 0.0939 - accuracy: 0.9648 - val_loss: 0.3314 - val_accuracy: 0.8542
Epoch 24/50
54/54 [==============================] - 46s 848ms/step - loss: 0.0770 - accuracy: 0.9742 - val_loss: 0.1303 - val_accuracy: 0.9427
Epoch 25/50
54/54 [==============================] - 46s 846ms/step - loss: 0.0805 - accuracy: 0.9701 - val_loss: 0.1726 - val_accuracy: 0.9375
Epoch 26/50
54/54 [==============================] - 46s 849ms/step - loss: 0.0701 - accuracy: 0.9759 - val_loss: 0.0793 - val_accuracy: 0.9688
Epoch 27/50
54/54 [==============================] - 46s 857ms/step - loss: 0.0860 - accuracy: 0.9677 - val_loss: 0.0530 - val_accuracy: 0.9792
Epoch 28/50
54/54 [==============================] - 46s 853ms/step - loss: 0.0373 - accuracy: 0.9859 - val_loss: 0.0288 - val_accuracy: 0.9792
Epoch 29/50
54/54 [==============================] - 46s 843ms/step - loss: 0.0506 - accuracy: 0.9800 - val_loss: 0.0456 - val_accuracy: 0.9896
Epoch 30/50
54/54 [==============================] - 46s 846ms/step - loss: 0.0338 - accuracy: 0.9883 - val_loss: 0.0274 - val_accuracy: 0.9896
Epoch 31/50
54/54 [==============================] - 46s 846ms/step - loss: 0.0400 - accuracy: 0.9859 - val_loss: 0.0494 - val_accuracy: 0.9792
Epoch 32/50
54/54 [==============================] - 46s 847ms/step - loss: 0.0355 - accuracy: 0.9824 - val_loss: 0.1104 - val_accuracy: 0.9531
Epoch 33/50
54/54 [==============================] - 46s 853ms/step - loss: 0.0761 - accuracy: 0.9718 - val_loss: 0.0464 - val_accuracy: 0.9740
Epoch 34/50
54/54 [==============================] - 46s 849ms/step - loss: 0.0338 - accuracy: 0.9883 - val_loss: 0.0391 - val_accuracy: 0.9844
Epoch 35/50
54/54 [==============================] - 46s 852ms/step - loss: 0.0302 - accuracy: 0.9906 - val_loss: 0.0561 - val_accuracy: 0.9844
Epoch 36/50
54/54 [==============================] - 46s 859ms/step - loss: 0.0570 - accuracy: 0.9783 - val_loss: 0.1028 - val_accuracy: 0.9740
Epoch 37/50
54/54 [==============================] - 46s 851ms/step - loss: 0.0486 - accuracy: 0.9830 - val_loss: 0.0309 - val_accuracy: 0.9896
Epoch 38/50
54/54 [==============================] - 46s 850ms/step - loss: 0.0821 - accuracy: 0.9701 - val_loss: 0.0282 - val_accuracy: 0.9844
Epoch 39/50
54/54 [==============================] - 46s 846ms/step - loss: 0.0646 - accuracy: 0.9707 - val_loss: 0.0689 - val_accuracy: 0.9688
Epoch 40/50
54/54 [==============================] - 46s 847ms/step - loss: 0.0438 - accuracy: 0.9830 - val_loss: 0.0485 - val_accuracy: 0.9792
Epoch 41/50
54/54 [==============================] - 46s 848ms/step - loss: 0.0688 - accuracy: 0.9777 - val_loss: 0.0488 - val_accuracy: 0.9792
Epoch 42/50
54/54 [==============================] - 46s 845ms/step - loss: 0.0353 - accuracy: 0.9883 - val_loss: 0.0373 - val_accuracy: 0.9896
Epoch 43/50
54/54 [==============================] - 46s 848ms/step - loss: 0.0388 - accuracy: 0.9883 - val_loss: 0.0561 - val_accuracy: 0.9688
Epoch 44/50
54/54 [==============================] - 46s 850ms/step - loss: 0.0195 - accuracy: 0.9918 - val_loss: 0.0477 - val_accuracy: 0.9792
Epoch 45/50
54/54 [==============================] - 46s 848ms/step - loss: 0.0452 - accuracy: 0.9836 - val_loss: 0.0352 - val_accuracy: 0.9896
Epoch 46/50
54/54 [==============================] - 46s 848ms/step - loss: 0.0215 - accuracy: 0.9912 - val_loss: 0.0173 - val_accuracy: 0.9948
Epoch 47/50
54/54 [==============================] - 46s 845ms/step - loss: 0.0357 - accuracy: 0.9894 - val_loss: 0.0537 - val_accuracy: 0.9792
Epoch 48/50
54/54 [==============================] - 46s 844ms/step - loss: 0.0316 - accuracy: 0.9871 - val_loss: 0.0486 - val_accuracy: 0.9844
Epoch 49/50
54/54 [==============================] - 46s 845ms/step - loss: 0.0100 - accuracy: 0.9959 - val_loss: 0.0569 - val_accuracy: 0.9740
Epoch 50/50
54/54 [==============================] - 46s 853ms/step - loss: 0.0336 - accuracy: 0.9883 - val_loss: 0.0651 - val_accuracy: 0.9792
In [27]:
scores = model.evaluate(test_set)

8/8 [==============================] - 2s 138ms/step - loss: 0.0527 - accuracy: 0.9805
In [28]:
scores
Out[28]:
[0.052731577306985855, 0.98046875]
In [29]:
 history.history.keys()
Out[29]:
dict_keys(['loss', 'accuracy', 'val_loss', 'val_accuracy'])
In [31]:
len(history.history['accuracy']) # equal to number of EPOCH
Out[31]:
50
In [32]:
accs = history.history['accuracy']
val_accs = history.history['val_accuracy']
losses = history.history['loss']
val_losses = history.history['val_loss']
In [36]:
# Training and Validation Accuracy Plot

plt.figure(figsize=(8,8))
plt.subplot(1,2,1)
plt.plot(range(EPOCHS), accs, label = 'Training Accuracy')
plt.plot(range(EPOCHS), val_accs, label = 'Validation Accuracy')
plt.legend(loc='lower right')
plt.title('Training and Validation Accuracy')

# Training loss Plot

#plt.figure(figsize=(8,8))
plt.subplot(1,2,2)
plt.plot(range(EPOCHS), losses, label = 'Training loss')
plt.plot(range(EPOCHS), val_losses, label = 'Validation loss')
plt.legend(loc='upper right')
plt.title('Training and Validation loss')

plt.show()
No description has been provided for this image
In [39]:
for images_batch, labels_batch in test_set.take(1):

    first_image = images_batch[0].numpy().astype('uint8')
    first_label = labels_batch[0].numpy()
 
    print("first image to predict")
    plt.imshow(first_image)
    print("actual label:", class_names[first_label])
    
    batch_prediction = model.predict(image_batch)
    #print(np.argmax(batch_prediction[0]))

    print('predicted label :', class_names[np.argmax(batch_prediction[0])] )

first image to predict
actual label: Potato___Early_blight
1/1 [==============================] - 0s 164ms/step
predicted label : Potato___Late_blight
No description has been provided for this image
In [ ]:
In [41]:
def predict(model, img):

    img_array = tf.keras.preprocessing.image.img_to_array(images[i].numpy())
    img_array = tf.expand_dims(img_array, 0) # create a batch

    predictions = model.predict(img_array)
    predicted_class = class_names[np.argmax(predictions[0])]
    confidence = round(100 * (np.max(predictions[0])), 2 )

    return predicted_class, confidence
In [44]:
plt.figure(figsize=(15, 15))
for images, labels in test_set.take(1):
    for i in range(9):
        ax = plt.subplot(3,3, i + 1)
        plt.imshow(images[i].numpy().astype("uint8"))
        predicted_class, confidence = predict(model, images[i].numpy())
        actual_class = class_names[labels[i]]

        plt.title(f"Actual: {actual_class}, \n Predicted: {predicted_class},\n Confidence: {confidence}%")
        
        plt.axis('off')
        

1/1 [==============================] - 0s 21ms/step
1/1 [==============================] - 0s 18ms/step
1/1 [==============================] - 0s 19ms/step
1/1 [==============================] - 0s 20ms/step
1/1 [==============================] - 0s 19ms/step
1/1 [==============================] - 0s 20ms/step
1/1 [==============================] - 0s 19ms/step
1/1 [==============================] - 0s 18ms/step
1/1 [==============================] - 0s 17ms/step
No description has been provided for this image
In [ ]:
# Save the models
In [ ]:
# Save the models
In [60]:
import os

model_version = 2#max([int(i) for i in os.listdir("../models") + [0]])+1
In [61]:
model.save(f"../models/{model_version}")

INFO:tensorflow:Assets written to: ../models/2/assets

INFO:tensorflow:Assets written to: ../models/2/assets
In [ ]: